In this case, one of the "issues" of machine learning is that the algorithm is not going to be any clearer than her headcanon. Of course, you could look at the training dataset (1 million pictures of the real world and 91,749 image of artworks by various artists according), but I do not think it will do you much good to explain the coloring choices that were made by the algorithm. And do not forget that other headcannons were used for the training of the algorithm:
We made several iterations: it had to match the ten pages worth of historical commentary about the paintings, in addition to Dr. Smola’s expert judgment as a result of studying Klimt’s artworks over several decades.
In the case of the art curator, at least you can ask her questions.